
EC-A8550JD4

CSR2-N72R3399

RS-F1SG2042

EC-ThorT5000

CORE-8550JD4

ICORE-1126BQ38

ICORE-1126BJQ38

CQ38W-1126B

EC-ThorT5000
· High-Performance RISC-V Processor K3
· Super Parallel Computing Capability
· Supports RISC-V Hardware Virtualization
· High-Level Defense Technology
· LPDDR5 Memory & UFS2.2 Storage
· 4K Hardware Video Encoding & Decoding
· Native support for Linux mainline kernel
· Industrial Metal Chassis with Heat Dissipation
Gemma
Qwen
Llama
Phi
Powered by SpacemiT Key Stone K3 chip with RISC-V homogeneous fusion architecture. Integrates a self-developed 8-core general-purpose big core X100 and an 8-core parallel AI core A100, delivering 130K DMIPS of general-purpose computing power and 60 TOPS of AI computing power. It can run large models with up to 30 billion parameters and supports rapid deployment of all types of AI algorithms and models. K3 natively supports Linux 7.0 mainline kernel and RISC-V hardware virtualization, features M/S/U three-level privilege architecture.
RISC-V Processor K3
Super Parallel Computing
RISC-V Hardware Virtualization
High-Level Defense Technology
32GB LPDDR5 Memory
Support 4K Video Codec
Native Linux mainline kernel
Industrial-Grade Metal Chassis
60 TOPS AI computing power with full support for the RVV vector instruction set, native FP8 inference, and multimodal algorithm acceleration. Capable of running 30B large language models and Mixture of Experts (MoE) models, with measured performance exceeding 10 Tokens/S@30B.
Widely applicable in industry fields such as edge computing, computing power services, data security, AI education, private deployment of large models, and self-service terminals.

AIBOX-K3 |
||
|
Basic Specifications |
SOC |
SpacemiT Key Stone K3 |
|
CPU |
8-core X100 64-bit RISC-V AI Processor, max main frequency 2.4GHz, 130KDMIPS general computing power, full RVA23 Profile support, single- core Specint2006>9.0/GHz, equivalent to ARM-A76 |
|
|
AI performance |
8-core A100 delivers 60TOPS AI computing power (SquareINT4), supporting BF16, FP16, FP8, INT8, INT4 and other data formats. Equipped with RVV vector instruction set, native FP8 inference and multi-modal algorithm acceleration, it supports 30B local models and MoE models, with inference performance exceeding 10Tokens/S@30B, and adapts to various AI algorithms and model deployment. |
|
|
GPU |
Integrated 3D graphics engine, supporting Vulkan 1.3, OpenCL 3.0, OpenGL ES 1.1/3.2, EGL 1.5 |
|
|
Codec |
Video decoding: 4K@120fps H.264/VP9 Video encoding: 4K@60fps H.264 |
|
|
RAM |
8GB/16GB/32GB LPDDR5 |
|
|
Storage |
128GB/256GB/512GB UFS2.2 |
|
|
Expansion storage |
1 × M.2 (PCIe3.0x4 lane, support for 2242/2280 specification PCIe NVMe SSD; located inside the computer) |
|
|
Power |
DC 12V/5A (5.5 × 2.1mm, supporting 9V~20V wide voltage input) |
|
|
OS |
Ubuntu 26.04, Buildroot Linux with QT GUI, Fedora, Bianbu OS, Deepin, OpenKylin, OpenHarmony 6.1, OpenEuler |
|
|
Size |
93.4mm × 93.4mm × 50.0mm |
|
|
Weight |
≈ 500g |
|
|
Environment |
Operating Temperature: -20℃~60℃, Storage Temperature: -20℃~70℃, Operating Humidity: 10%~90%RH (No condensation) |
|
|
Interface Specifications |
Ethernet |
2 × Gigabit Ethernet (1000Mbps/RJ45) |
|
Video output |
1 × HDMI2.0 (Up to 4K@60Hz) |
|
|
USB |
2 × USB3.0 (Max: 1A), 1 × USB3.0 DRD (Type-C, includes USB2.0 OTG) |
|
|
Console |
1 × Console (Debug serial) |
|
|
Button |
1 × Power, 1 × Boot |
|
Firefly team, with over 20 years of experience in product design, research and development, and production, provides you with services such as hardware, software, complete machine customization, and OEM server.